Which super-resolution algorithm is proper for Farsi text image sequences

Abstract

In this paper we propose a new algorithm for super-resolution of Farsi text image sequences. Our algorithm contains three main steps as prior super-resolution algorithms; registration, reconstruction, and restoration. Due to special properties of Farsi texts such as appearance of dots in alphabet, selecting a proper super-resolution algorithm, especially in presence of noise, is more important. We propose an algorithm with an accurate sub-pixel registration and IBP reconstruction that reconstructs a high resolution image from a set of noisy low resolution observations. In restoration step we have exploited NLM algorithm to overcome image noise. We test our algorithm on synthetic and real data. Both quantitative and qualitative results show outperformance of our algorithm.